The AI battle during the Spring Festival season has now largely concluded. Leading tech companies ramped up their activities before the festival, and I planned to review their performances afterward, ranking each product from solid to spectacular.

There were rumors that DeepSeek would release R2, which I personally expected to be the most promising contender for the 'solid' title. Unfortunately, R2 followed suit with its usual pattern of delays.

If we take R1's performance from a year ago as the benchmark for 'solid,' then clearly no product this year has met that standard.

However, the biggest difference between this year and last is that while DeepSeek stood out alone last year, this year has seen a more vibrant landscape with many active contenders. Both the technical and marketing aspects are worth discussing, offering a much richer pool of material for evaluation.

Below is a brief review of the AI battle performances of the BAT trio during the Spring Festival season.

ByteDance Defends, Alibaba Attacks

Judging by overall performance, both ByteDance and Alibaba qualify as top-tier contenders.

When it was officially announced that ByteDance had secured the CCTV Spring Festival Gala partnership, the narrative emphasized Volcano Engine as the main player, with Doubao playing a supporting role. However, in reality, Doubao ended up receiving the majority of the exposure resources.

According to Doubao's data, users engaged in a total of 1.9 billion AI interactions on New Year's Eve, generating over 50 million festive avatars and 100 million New Year greetings.

These figures can only be described as fairly average, as past collaborations between internet companies and the CCTV Spring Festival Gala have often resulted in interaction counts in the tens of billions. However, ByteDance's early securing of the CCTV Spring Festival Gala partnership itself represents top-tier strategic foresight and positioning.

While enthusiasm among major internet companies for the CCTV Spring Festival Gala has waned in recent years, it remains a significant channel for product promotion.

When ByteDance finalized its CCTV Spring Festival Gala partnership, the intensity of AI competition was not as high, which may explain why Volcano Engine took the lead at the time. By the time the Spring Festival AI battle began, such promotional channels became especially crucial.

ByteDance has Douyin, so Doubao is not short of promotional channels, but Alibaba is. Therefore, having the CCTV Spring Festival Gala in ByteDance's hands constitutes a strategic check on Alibaba.

Qianwen ultimately settled for sponsoring the Spring Festival Galas of four provincial satellite TV stations, an outcome likely inferior in both impact and cost-effectiveness compared to the CCTV Spring Festival Gala.

However, Alibaba had limited options for spending its budget, as ByteDance and Tencent are both rivals, collectively controlling most of the internet traffic.

Another reason for ranking ByteDance as top-tier is that Seedance 2.0 achieved state-of-the-art (SOTA) performance in video generation.

This model has become highly popular among creators and media circles, even leading to situations where foreign entities sought 'reverse invitation codes' from China. However, KOL opinions should be taken with a grain of salt, as these same individuals previously hyped Kling and Veo using similar tactics.

Moreover, objectively speaking, video generation models remain relatively niche and have not truly broken into the mainstream. The core indicator for judging technological strength still lies in language models.

Alibaba's Qianwen appears to be the biggest spender this year.

Ranking Alibaba as top-tier is not (solely) because I've had a few cups of milk tea from Qianwen's promotions. Rather, while others were busy with mundane New Year greetings, only Alibaba was genuinely considering how to use AI to fundamentally transform the existing internet operating model.

From a pioneering perspective, Qianwen's ordering gameplay echoes Alibaba's earlier ventures with Taobao and Alipay, representing a high-risk but potentially high-reward strategic bet. Although the current Qianwen ordering experience is still relatively basic, it can continuously improve as model capabilities advance.

Of course, the extent of improvement and whether it can ultimately provide users with value surpassing traditional models remains highly uncertain.

Many are skeptical about the AI shopping model, which is understandable. However, it's important not to directly link this pessimism to Qianwen's fate. Qianwen's Spring Festival hospitality plan is a marketing campaign, and its worth depends on how many users it attracts and retains.

Given that Qianwen started late compared to its rivals, using aggressive tactics to rapidly raise its profile is necessary, especially from a team morale perspective. Moreover, Alibaba's ability to integrate various business lines for AI shopping in such a short time demonstrates commendable organizational capability and combat effectiveness.

Alibaba's mistake, in my view, was missing out on the CCTV Spring Festival Gala partnership, which could not be compensated for by signing with four provincial satellite TV stations. The new version of the Qianwen app launched its public beta on November 17. Public beta implies that functions are not yet complete and the experience needs refinement.

Volcano Engine announced its CCTV Spring Festival Gala partnership on December 29, and the process from project initiation, negotiation, to finalizing the cooperation would have taken much longer. Qianwen would have found it difficult to meet this timeline, likely focusing its efforts on functional completion and updates instead.

Overall, ByteDance defended skillfully, while Alibaba attacked decisively, making them evenly matched opponents, hence both receiving top-tier evaluations.

Tencent Yuanbao: Misjudgment in AI Social

Tencent's performance during this Spring Festival season pales in comparison to BA, and I would personally rate it no higher than NPC level.

According to AppGrowing estimates, Yuanbao's annual investment in user acquisition reached 15 billion yuan in 2025, with 5.763 billion yuan spent in the third quarter alone. However, QuestMobile's data shows Yuanbao's monthly active users in September were only 32 million. No wonder Pony Ma remarked that it would be better to save marketing expenses and distribute red packets to users instead, as Yuanbao's massive spending yielded limited results.

Yuanbao had two major moves during the Spring Festival season: 1 billion yuan in red packets and Yuanbao Pai. Red packets are a conventional means of user acquisition and engagement, not particularly noteworthy in themselves. However, the WeChat ecosystem's green light for Yuanbao's user acquisition efforts, with widespread screen-spamming messages significantly impacting user experience, is an undeniable fact.

The official's subsequent hesitation, awkwardness, and wavering in handling user acquisition for Yuanbao and Qianwen make it difficult for outsiders to believe their stated commitment to 'user experience first, treating everyone equally.'

Yuanbao Pai represents a different issue. Social networking is Tencent's strongest foundation, and the company is widely regarded as product-centric. However, when Tencent truly served up Yuanbao Pai under the banner of 'AI social,' it was hard to swallow, leaving only endless disappointment from unmet expectations.

The only positive aspect is that, without leveraging the momentum of WeChat and QQ, the fact that Tencent itself could only achieve such results with AI social suggests that Tencent's social moat in the AI era is indeed as formidable as imagined.

On February 23, the App Store's free download chart showed that Tencent Yuanbao had fallen out of the top 30. Meanwhile, ByteDance's Doubao, Alibaba's Qianwen, and Ant Group's Afu remained in the top five. On the other hand, Yuanbao still has not addressed the issue of insufficient foundational model capabilities.

A recent report by LatePost mentioned that before the Spring Festival, Yuanbao was simultaneously developing the social feature 'Yuanbao Pai' while awaiting access to DeepSeek's new model expected to be released during the festival.

However, enhancing foundational model R&D is an imperative issue. Relying on open-source models is only a short-term emergency solution for a company of Tencent's scale.

Dianchang summarized information from Tencent's annual meeting, mentioning that Martin believes Yuanbao's poor user retention and growth are fundamentally due to the inadequacy of the Hunyuan large model, and DeepSeek is unlikely to significantly aid Tencent in specific training.

Late last year, Tencent announced the official joining of Yao Shunyu as Chief AI Scientist, also heading the AI Infra and Large Language Model departments. Management has high hopes for Yao and has granted him significant authority.

Yao's top priority after taking over Tencent's AI R&D is clearly to enhance foundational model capabilities and deliver a competitive product that can match rivals' leading models. Based on his professional background, during his doctoral studies and time at OpenAI, Yao mainly focused on intelligent agents, with seemingly little experience in training foundational large models.

Given Yuanbao's obvious weaker performance compared to its two rivals during the Spring Festival, Yao may need to prove his capabilities and value more quickly. Fortunately, Yao may not necessarily have to achieve this by helping Yuanbao turn the tide.

The WeChat Agent is a definite plan, perfectly aligning with Yao's expertise and representing a much larger opportunity than Yuanbao.

Helping Yuanbao make a comeback is difficult, as its rivals are ByteDance and Alibaba. Developing the WeChat Agent is also challenging, primarily in finding a reasonable paradigm that convinces Allen Zhang and Pony Ma that the timing is right.

However, once the decision is made, achieving results becomes an expected outcome.